Research map / Vision-language and multimodal / Robot action and navigation
PASG: A Closed-Loop Framework for Automated Geometric Primitive Extraction and Semantic Anchoring in Robotic Manipulation
Zhihao Zhu, Yifan Zheng, Siyu Pan, Yaohui Jin, Yao Mu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| LoRA fine-tune on Robocasa-PA questions (Qwen2.5VL-PA) | Qwen2.5-VL · Robocasa-PA in-distribution | overall accuracy: 43.91 -> 77.79 | Improved |
| LoRA fine-tune on Robocasa-PA questions | Qwen2.5-VL · Robocasa-PA out-of-distribution | overall accuracy: 43.33 -> 79.69 | Improved |
| replace human annotation with PASG automatic annotation | RoboTwin policy · RoboTwin (6 tasks) | average success rate: 84.67 -> 77.83 | Hurt |
| replace human annotation with PASG annotation | RoboTwin policy · Block Hammer Beat | success rate: 79.0 -> 82.0 | Improved |
| replace human annotation with PASG annotation | RoboTwin policy · Empty Cup Place | success rate: 73.0 -> 76.0 | Improved |
| replace human annotation with PASG annotation | RoboTwin policy · Dual Bottles Pick | success rate: 95.0 -> 70.0 | Hurt |
| replace human annotation with PASG annotation | RoboTwin policy · Messy Shoe Place | success rate: 83.0 -> 69.0 | Hurt |
| fine-tune on 5% of training data | Qwen2.5-VL · Robocasa-PA | accuracy: approx. 10% absolute, 20.6% relative | Improved |
| fine-tune on 10% of training data instead of 5% | Qwen2.5-VL · Robocasa-PA | accuracy: 20% absolute, 41.12% relative | Improved |
| add self-refine matching loop with finer segmentation granularity | PASG · Own dataset | matching success rate: 98% | Improved |
Closest papers by meaning
- OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints (CVPR 2025 · similarity 0.90)
- SoFar: Language-Grounded Orientation Bridges Spatial Reasoning and Object Manipulation (NeurIPS 2025 · similarity 0.90)
- Localizing, Structuring, and Rendering: Bridging 3D and 2D Vision-Language-Action Models for Robotic Manipulation (CVPR 2026 · similarity 0.90)
- RealVLG-R1: A Large-Scale Real-World Visual-Language Grounding Benchmark for Robotic Perception and Manipulation (CVPR 2026 · similarity 0.89)
- RoboGround: Robotic Manipulation with Grounded Vision-Language Priors (CVPR 2025 · similarity 0.88)
- Seeing Across Views: Benchmarking Spatial Reasoning of Vision-Language Models in Robotic Scenes (ICLR 2026 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
